My sharpening workflow for base ISO mages with the new features in RT (aka extreme pixel peeping)


(Ingo Weyrich) #21

You can already try them out (except the automated calculation of dual-demosaic contrast threshold).


(Stefan Chirila) #22

@heckflosse where would I download the appimage ? I run debian linux


(Ingo Weyrich) #23

RawTherapee-dev-5.4-412-gb626b4e9.AppImage


(Stefan Chirila) #24

@heckflosse sweet will try out :smiley:


(Ingo Weyrich) #25

@stefan.chirila Please report your findings. Any suggestion to improve the new stuff is welcome.


(Sebastien Guyader) #26

Info, is your current automated threshold code in a branch I could try?


(Ingo Weyrich) #27

Sébastien, not yet.


(Ingo Weyrich) #28

That was nonsense. Bright parts are taken from the sharpened image while dark parts are taken from the unsharpened image.

Using the procedure @hholt mentioned here, the mask preview can also be used to tweak the contrast threshold for AMaZE+VNG4 or xtrans 4-pass demosaic.


(Samuel Chia) #29

@heckflosse Hi Ingo, I do apologize for not keeping up with my thread on demosaicing and diagonal artifacts with AMaZE to the end. Life got in the way and I must also admit that some of the technical discussions of implementation of the RCD algorithm went above my head. I got an email update of new topics and noticed with great interest, this new thread. I see that you have combined AMaZE and VNG4. I also saw that RCD has been implemented too at some point? I’m not entirely sure if the diagonal artifact issue still remains with the current implementation of RCD.

I’m asking as I can still see what looks like residual diagonal artifacts from the AMaZE demosaicing. My last test with the RCD algorithm way back I recall seemed to handle noise pretty well and not result in the blocky line artifacts which AMaZE and DCB suffer from. Luis mentioned it would be possible to tune the RCD code to not produce artifacts in noisy parts of the image at the cost of fine detail. Also mentioned was that RCD does not deal with moire issues, but I did not test yet. So maybe RCD is not necessarily better overall.

Another good combination might be DCB + VNG4. I’m glad you picked VNG4 as the interpolator for noisy regions of the image. Of all the current interpolators I think it produces the smoothest result. I currently do a blend of DCB and VNG4 by using the “Find Edges” filter in Photoshop to create a mask, so the DCB rendered image is applied in the high contrast edges of a given image, while the smooth regions have VNG4 applied.

Regarding the masking, what do you think about having the high frequency detail ultilise AMaZE/DCB/RCD, and regions of low frequency detail to be taken from VNG4? Rather than using luminosity as the determinator for which demosaicing to apply. But perhaps I have misunderstood this part as I see the mask in grafik.jpeg shows that only the edges of the tower to be bright, while all the smooth regions are dark.


(Ingo Weyrich) #30

Should be no problem to adapt the current code for other combinations of demosaicers. I already use that to combine xtrans 3-pass and xtrans-fast in the new xtrans 4-pass.

You did. The method is based on contrast, not on luminosity


(Stefan Chirila) #31

@heckflosse I downloaded it and it runs on my computer relatively well …a little bit laggy but nothing I can’t deal with (still not the instant response I’d get from the old gtk2 version; though I know this new one has tons of new features which likely is using more processor). Lovely work on the new features; and I’ll upload a quick edit I did with the sharpening technique on a street photo I took on my trip overseas which I just got back from. The photo was taken through a car window while driving, so I’m really happy with the level of contrast enhancement the technique gives. I don’t have the pp3 on hand right at the moment, but I basically did an auto levels on it, then BW conversion to Orthochromatic (because that’s my favourite look).

Update:
I have found I enjoy this type of edit a lot; a relatively low contrast BW, with very enhanced local contrast and details. It reminds me of the old BWs, such as work by James Nachtwey (his images) By the way his documentary War Photographer (and on Wikipedia) is a must if you enjoy seeing photography put to purposeful use to positively change the world.


(Ingo Weyrich) #32

@samuelchia I will answer more detailed tomorrow.


(Gabriele ) #33

Hi Ingo,
this make me clearer the thing

Thanks


(Ingo Weyrich) #34

Yes, I got this idea while working on a contrast based threshold to sharpen only regions with contrast > threshold while not sharpening regions with contrast <= threshold (with a special blend function to smooth transitions between both areas). The new sharpening threshold is available in dev versions of RT.

For RCD @cuniek is the specialist

At least in flat regions AMaZE, DCB and RCD show block line artifacts when the output is sharpened. For that reason I use VNG4 for the flat regions in combined demosaic.


(Samuel Chia) #35

Thanks @heckflosse for the explanations and clarification that the masking is indeed contrast-based and not luminance-based!

That’s great, my motivations are similar to yours then. The tonally smooth regions just look too funky with the block line artifacts. I just did a thorough evaluation last night with AMaZE, DCB and RCD, and I can confirm that I see the same as you do, they all suffer from the blocky artifacts in smooth areas. I also noticed that RCD is still making diagonal line artifacts in certain areas, while DCB is the only interpolator that is diagonal-line artifact-free. For general usage, I think DCB is still the better interpolator overall. If one has problems with moire, RCD is a poor choice and AMaZE is the best. Unfortunately, AMaZE continues to have the strong diagonal ringing artifacts. RCD turns out to be excellent for avoiding false colour artifacts in the implementation in the stable release of V5.4, and for round edges like stars, it does an excellent job, and can be a bit to a lot better than DCB or AMaZE, smoother edges and less colour fringing. So for astrophotography, RCD+VNG4 would be ideal. For general use, DCB+VNG4 is great, and for regions with moire, one could blend in AMaZE for the problem areas.

There is no one ideal demosaicing interpolator, each have their own strengths and issues. For this reason I think it would be great to have the DCB and RCD with VNG4 combos in addition to AMaZE. I’ve not downloaded and compiled (not sure how to compile) this dev version of RT, and I don’t see the demosaic panel in the screenshots of the first post. How does one select the demosaic method? Is it a drop down menu where you can choose from any demosaic interpolator currently available and thus make any combo you like? Or only one AMaZE+VNG4 choice (and the X-trans method) is available?


(Peter Lavender) #36

Hi @samuelchia,

The demosaicing methods are found under the little square that looks like a chequed flag:

Screenshot%20from%202018-06-23%2015-13-19

Rawpedia does a great job detailing how to compile Rawtherapee on all Operating Systems: https://rawpedia.rawtherapee.com/Main_Page#Compiling

I pretty much only run the latest dev branch with my editing session starting with opening a terminal and doing a git pull and then compile.

I’ve processed the below photo using my normal processing method and changing the Demosaicing to Amaze+VNG4. I think it’s not too bad for an ISO2500 image. Happy to provide the raw file too if anyone wants to try it out.

And similar with just Amaze:

Uploading…


(Ingo Weyrich) #37

I tested with some astrophotography raws I found in the web and I agree that for single astro raw files RCD is better than AMaZE. But isn’t the usual procedure to stack a lot of astro raw files to avoid usual demosaicers?


(Ingo Weyrich) #38

For a single astrophotography raw file the currently available AMaZE+VNG4 is much better than plain AMaZE. Look at this screenshot, left is AMaZE, right is AMaZE+VNG4

The difference between plain AMaZE and plain RCD is less obvious. Left is plain AMaZE, right is plain RCD


(nosle) #39

I quite like this feature for portraits. I frequently mask sharpening (in other software) to avoid noise in the background blur. This feature has a similar effect without having to draw a mask. Nice one!


(Gimbal Lock) #40

I’m no expert in astrophotography but when I do try it I have to convert the raws to tiff before stacking, mainly because my stacking software of choice doesn’t support Pentax raws natively. Which means I always end up wondering which settings (white balance and such) is the best to use when converting to tiff in order to loose as little information as possible.